Cardiometabolic Monitoring and Sociodemographic and Clinical Characteristics of Youths Prescribed Antipsychotic Medications

This was a serial cross-sectional analysis of electronic health record (EHR) data for 4,568 youths initiating second-generation antipsychotic medication treatment during 2013–2017 in Kaiser Permanente Northern California (KPNC), a large integrated health care system serving >4 million socioeconomically and clinically diverse patients. KPNC patients are representative of the Northern California population, except for those at the lowest and highest incomes (19). Patients may enroll in KPNC through health insurance plans, including employer-based coverage, Medicare, Medicaid, and health insurance exchanges. The KPNC Institutional Review Board approved the study, with a waiver of consent. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

Youths, ages 10–21 years, were included if they had two prescription fills of one of the following second-generation antipsychotic medications within a 12-month window during 2013–2017: aripiprazole, quetiapine, risperidone, olanzapine, clozapine, ziprasidone, asenapine, iloperidone, lurasidone, or paliperidone. The first prescription date was considered the index date. Additional inclusion criteria were no pregnancy and continuous KPNC enrollment and drug coverage for the year before and 2 years after the index date, allowing for brief lapses in enrollment (i.e., no more than a total of 3 months). To constrain the cohort to youths with new second-generation antipsychotic prescriptions, we excluded youths with any antipsychotic medication fill during the year before the index date. The total observation range for this study was January 1, 2013, through December 31, 2019.

Patient sociodemographic and clinical characteristics, health care utilization, and cardiometabolic risk factor monitoring were extracted from the EHR. Racial-ethnic categories were White; Black or African American (Black); Hispanic or Latino (Latino); Asian American, Native Hawaiian, or Pacific Islander (AANHPI); and another or unknown race-ethnicity. We also assessed age, gender, Medicaid insurance, and median annual neighborhood income (income was based on census tract data for the youth’s residential address; ≤$50,000 was considered low income).

Mental and behavioral health diagnoses were considered present at baseline if documented in the EHR 1 year before and after the index date (2 years surrounding the index date; see Table S1 in the online supplement to this article for the list of ICD-9 and ICD-10 codes used for classifying diagnostic groups). We constructed a hierarchical variable capturing presence or absence of serious mental illness according to three categories: any psychotic disorder, any bipolar disorder, and neither of the above. We additionally measured the following four dichotomous indicators representing presence or absence of nonbipolar affective disorder (i.e., major depression or anxiety disorder): substance use disorder; attention-deficit hyperactivity disorder; developmental, intellectual, or disruptive disorder; and other behavioral health diagnosis—these diagnostic categories were not mutually exclusive. By using pharmacy dispensation records, we identified an index second-generation antipsychotic prescription on the basis of two documented fills of a given antipsychotic type. Obesity status was determined by using age-adjusted body mass index (BMI) percentiles during the year before the index date, with the 95th percentile indicating obesity for youths ages 10–19 and a BMI of ≥30 indicating obesity for youths ages 20–21 (coded as observed or not observed). Presence of a comorbid cardiometabolic condition was defined as any ICD-10 diagnostic codes or laboratory results (see Table S2 in the online supplement) indicating prediabetes, diabetes, hypertension, or dyslipidemia. Behavioral health (e.g., psychiatry or addiction medicine) and primary care (e.g., internal or family medicine, obstetrics, gynecology, or women’s health) utilization were for the year after antipsychotic initiation.

The primary outcome was presence of cardiometabolic monitoring laboratory results, defined as having both glycemic and lipid tests performed within 2 years following the index date (follow-up). We selected a 2-year follow-up window to include laboratory results occurring any time during the calendar year after the index date. We also assessed monitoring rates within 1 year after the index date. Glycemic testing was defined as a fasting glucose serum or glycosylated hemoglobin test. Lipid testing was defined as a triglyceride or cholesterol test.

We used chi-square statistics to describe rates of glycemic and lipid testing and to assess differences by sociodemographic characteristics. To assess time trends in monitoring rates, we used a modified Poisson regression framework to estimate risk ratios (20, 21). We first estimated a generalized linear model including the year of antipsychotic medication initiation, sociodemographic and clinical characteristics, and health care utilization. To account for effects related to geographic or site-specific intercorrelation, we specified a generalized estimating equations model including the same variables but additionally accounting for clustering by medical facility service area. We also assessed potential sociodemographic moderators of the time effect on monitoring by estimating five additional generalized estimating equations models to examine interactions between year of antipsychotic initiation and each sociodemographic characteristic: gender, age (continuous), race-ethnicity, median neighborhood income, and insurance type.

Because of the data set size and multiple comparisons, we assessed statistical significance by using a conservative alpha of 0.002 and 99.8% CIs, consistent with a Bonferroni correction for 29 variables used in the regression models. Analyses were conducted in Stata, version 15.1.

For the 4,568 youths, the mean±SD age was 17.0±3.0 years; 52% of the participants were male; 50% were White, 22% Latino, 13% Black, and 12% AANHPI; 20% lived in a low-income neighborhood; and 17% had Medicaid insurance (Table 1). Males and females significantly differed in age distribution, behavioral health diagnoses, index second-generation antipsychotic medication type, and use of behavioral health and primary care services.

On average, 54% of youths completed cardiometabolic monitoring within 2 years of antipsychotic initiation, with similar rates for glycemic and lipid testing; 62% of youths received at least one of the test types (Table 2). Bivariate analyses showed that monitoring rates differed by gender, age, and race-ethnicity but not by neighborhood income or insurance type. Glycemic testing was less common among males than females. Monitoring completion significantly declined with age: only 48% of youths ages 19–21 years completed monitoring of both tests, compared with 64% of those ages 10–12. The racial-ethnic groups that were least likely to complete monitoring were Black and White (51% with both glycemic and lipid tests). AANHPI and Latino youths were most likely to complete monitoring (61% and 58%, respectively, with both test types). When the follow-up window was constrained to 1 year, 41% of youths were screened; sociodemographic patterns closely resembled those seen for monitoring at 2 years. Most youths (85%) had a BMI documented at baseline.

A generalized linear model adjusted for sociodemographic and clinical characteristics and utilization of behavioral health and primary care services (data not shown) indicated continued statistically significant age and race-ethnicity associations with completion of monitoring. Specifically, youths ages 19–21 years were less likely to receive monitoring than youths ages 10–12 (risk ratio [RR]=0.85, 99.8% CI=0.73–0.99, p=0.001; data not shown), and Latino and AANHPI youths were more likely to receive monitoring than White youths (e.g., for Latino youths, RR=1.11, 99.8% CI=1.01–1.23, p=0.001; data not shown). However, racial-ethnic associations were no longer significant when the model was adjusted for clustering by medical facility service area (Table 3).

Monitoring rates increased 5% annually (RR=1.05, p=0.001), as shown by the fully adjusted model that also accounted for clustering by medical facility service area (Table 3). In moderation analyses, monitoring time trends did not differ by sociodemographic factors, such as gender or race-ethnicity (data not shown).

Cardiometabolic monitoring was 23% more likely for youths with a psychotic disorder, compared with those without a serious mental illness (RR=1.23, p<0.001) (Table 3, Figure 1). Monitoring was also positively associated with risperidone versus quetiapine prescription (RR=1.17, p<0.001) and with observed obesity (RR=1.13, p=0.001). Presence of a comorbid cardiometabolic condition was not significantly associated with monitoring (RR=1.06), nor did this association become significant when obesity was removed from the model (RR=1.09; data not shown). Youths who visited primary care during the follow-up were more likely to receive monitoring than those who did not visit primary care, as were youths who had quarterly or more visits to behavioral health services (e.g., RR=1.47 vs. youths with no behavioral health visits, p<0.001). Monitoring was 19% less likely for youths with a substance use disorder (RR=0.81 vs. no substance use disorder, p<0.001). When the follow-up was constrained to 1 year, rates and p values were similar to those found in the 2-year outcome model.